Wraps Optuna to tune hyperparameters of a registered MOT tracker. Uses the tracker's ``search_space`` ClassVar to define the Optuna parameter

Metadata about a frame source used to drive progress display. Attributes: source_type: Kind of source (`video`, `image_dir`, `webcam`,

Tracklet for the BoT-SORT tracker. Uses ``XCYCWHStateEstimator`` (center + width/height) by default, mirroring the original BoT-SORT Kalman f

Context-manager that renders a single live progress line. Args: source_info: Source metadata returned by `_classify_source`. cons

Cascaded-Buffered IoU (C-BIoU) tracker. Implements the matching strategy from Yang et al., *Hard To Track Objects with Irregular Motions and

BoT-SORT-style multi-object tracker (IoU association + optional CMC). The tracker maintains a list of active tracks (Kalman-filter-based) an

Buffered Intersection over Union. Computes IoU after expanding each box by a configurable relative margin around its center: - ``x1' = x

ByteTrack operates online by processing all detector outputs, categorizing them by confidence thresholds to enable a two-stage association process

Standard Intersection over Union. Computes the ratio of the intersection area to the union area for every pair of boxes. Values range from 0

In SORT, object tracking begins with high-confidence detections fed into a Kalman filter framework assuming uniform motion for state prediction ac

Result for a single sequence evaluation. Attributes: sequence: Name of the sequence. CLEAR: CLEAR metrics for this sequence, or `

Context manager for MOT format file writing.

Configuration for camera motion compensation (CMC). The CMC module estimates a global 2D affine transform `H` (2x3) between consecutive

No-op transformation where absolute and relative coordinates are identical. Used for the first frame (before any camera motion is detected) or wh

Generic Kalman filter implementation. A standard linear Kalman filter for state estimation. This is a clean, general-purpose implementation t

Context manager for lazy video file writing.

Result for multi-sequence evaluation. Attributes: sequences: Dictionary mapping sequence names to their results. aggregate: Combi

Camera motion compensation estimator. Estimates a global 2D affine transform H (2x3) between consecutive frames and provides helpers to

Generalized Intersection over Union (Rezatofighi et al., 2019). Extends standard IoU by penalizing the empty area within the smallest enclosi

Full perspective transformation using a 3x3 homography matrix. Supports rotation, translation, scaling, and perspective changes. This is the

Tracklet for OC-SORT tracker with ORU (Observation-centric Re-Update). Manages a single tracked object with Kalman filter state estimation. I

Detection data for a single frame from a MOT format file. Attributes: ids: Track IDs for each detection. Shape `(N,)` where N is number

Test-double IoU that records non-empty compute calls.

Complete Intersection over Union (Zheng et al., 2019). Builds on **DIoU** by adding a penalty for mismatched aspect ratio between boxes (via

CLEAR metrics with TrackEval-compatible field names. Float metrics are stored as fractions (0-1 range), not percentages. The values follow the ori

Distance Intersection over Union (Zheng et al., 2019). Extends IoU by penalizing the normalized Euclidean distance between bounding-box cente

Supported benchmark tracking datasets. Attributes: MOT17: Pedestrian tracking with crowded scenes and frequent occlusions. St

HOTA metrics with TrackEval-compatible field names. HOTA evaluates both detection quality and association quality. Float metrics are stored as fra

Identity metrics with TrackEval-compatible field names. Identity metrics measure global ID consistency using an optimal one-to-one assignment betw

OC-SORT enhances traditional SORT by shifting to an observation-centric paradigm, using detections to correct Kalman filter errors accumulated dur

Holds metadata for a single tracker parameter. Stores the type, default value, and description extracted from the tracker's __init__ signatur

Holds a tracker class and its extracted parameter metadata. Used by the CLI to discover available trackers and their configurable options wit

Tracker parameter mapping with CLI-only filtering for IoU metrics.

Context manager for OpenCV display window with resizable output.

Prepared sequence data ready for metric evaluation. This dataclass contains all data needed by CLEAR, HOTA, and Identity metrics. IDs are rem

Abstract base for IoU similarity metrics used in tracker association. Subclasses implement a specific Intersection over Union variant (e.g. s

Abstract Kalman filter with a specific bounding box state representation. Wraps a `KalmanFilter` and provides a unified interface for boundin

Abstract tracker with auto-registration via tracker_id class variable. Subclasses that define `tracker_id` are automatically registered and b

Abstract base class for all tracker-specific tracklets. Provides common interface and attributes for tracklet management.

Abstract base class for coordinate transformations. Subclasses implement specific transformation types that convert points between absolute (

Downloadable asset types within a dataset split. Attributes: FRAMES: Raw video frames as individual image files. ANNOTATIONS: Gro

Available dataset splits. Attributes: TRAIN: Training split. VAL: Validation split. TEST: Test split.

Draws object trajectories with camera motion compensation. This annotator maintains a history of object positions in world coordinates and dr

Estimates camera motion between consecutive video frames. Uses sparse optical flow (Lucas-Kanade) to track feature points and computes the ge

Kalman filter state representation for bounding boxes. Attributes: XCYCSR: Center-based representation with 7 state variables:

Multi-sequence aggregation of HOTA metrics.

Multi-sequence aggregation of Identity metrics.

Verify behavior of Buffered IoU.

BoT-SORT camera motion compensation integration.

BoT-SORT-specific lifecycle behavior.

Intersection-over-area computation.

Pairwise intersection-over-union computation.

BIoU should associate near-miss detections that plain IoU would miss.

With buffer_ratio=0, BIoU recovers IoU; C-BIoU should match BoT-SORT (no CMC).

Verify mathematical properties of CIoU.

`CMC.apply_batch` against center-state tracklets. All tests build default `BoTSORTTracklet(bbox)` instances (center-state estimator) and veri

Adversarial H inputs to CMC.apply_batch. Covers wrong-shape H (should raise) and non-finite H values (should propagate to state without raisi

Direct unit tests on `CMC.warp_xyxy_corners(x1, y1, x2, y2, R, t=None)`. Each test passes raw 1-D NumPy arrays for the four corner channels and a

CMC.estimate() contract — parametrised over every supported method. All tests instantiate CMC via `_make_cmc(method)` (skips when the method is

Per-sequence HOTA metric computation (DetA, AssA, LocA, HOTA).

Per-sequence Identity metric computation (IDTP, IDFN, IDFP, IDF1).

Verify mathematical properties of DIoU.

Pin behavior of BaseIoU.compute on edge-case inputs.

Verify BaseIoU.compute handles empty inputs for all subclasses.

MOT sequence evaluation: single-metric, multi-metric, and output formats.

Tests for _fuse_score — IoU similarity fused with detection scores.

Verify mathematical properties of GIoU.

Tests for _get_alive_tracklets — tracklet lifecycle filtering.

_get_associated_indices returns sorted unmatched lists across all trackers.

Compare IoU variants against reference implementations.

GT preprocessing must follow TrackEval's class-based distractor handling. TrackEval (`mot_challenge_2d_box.py`) scores only the pedestrian class

Output of --list.